Practical data analysis relies on the ability to count observations of objectssuccinctly and efficiently. Unfortunately the space usage of an exact estimator grows with the size of the a priori set from which objects are drawn while the time required to maintain such an estimator grows with the size of the data set. We present static and on-line approximation schemes that avoid these limitations when approximate frequency estimates are acceptable. Our Log-Frequency Sketch extends the approximate counting algorithm of Morris [Morris1978] to estimate frequencies with bounded relative error via a single pass over a data set. It uses constant space per object when the frequencies follow a power law and can be maintained in constant time per observation. We give an (epsilon, delta)-approximation scheme which we verify empirically on a large natural language data set where, for instance, 95 percent of frequencies are estimated with relative error less than 0.25 using fewer than 11 bits per object in the static case and 15 bits per object on-line.

Detecting abnormal activities from sensor readings is an important research problem in activity recognition. A number of different algorithms have been proposed in the past to tackle this problem. Many of the previous state-based approaches suffer from the problem of failing to decide the appropriate number of states, which are difficult to find through a trial and-error approach, in real-world applications. In this paper, we propose an accurate and flexible framework for abnormal activity recognition from sensor readings that involves less human tuning of model parameters. Our approach first applies a Hierarchical Dirichlet Process Hidden Markov Model (HDP-HMM), which supports an infinite number of states, to automatically find an appropriate number of states. We incorporate a Fisher Kernel into the One-Class Support Vector Machine (OCSVM) model to filter out the activities that are likely to be normal. Finally, we derive an abnormal activity model from the normal activity models to reduce false positive rate in an unsupervised manner. Our main contribution is that our proposed HDP-HMM models can decide the appropriate number of states automatically, and that by incorporating a Fisher Kernel into the OCSVM model, we can combine the advantages from generative model and discriminative model. We demonstrate the effectiveness of our approach by using several real-world datasets to test our algorithm’s performance.

Bootstrap Voting Experts (BVE) is an extension to the Voting Experts algorithm for unsupervised chunking of sequences. BVE generates a series of segmentations, each of which incorporates knowledge gained from the previous segmentation. We show that this method of bootstrapping improves the performance of Voting Experts in a variety of unsupervised word segmentation scenarios, and generally improves both precision and recall of the algorithm. We also show that Minimum Description Length (MDL) can be used to choose nearly optimal parameters for Voting Experts in an unsupervised manner.

Here novelty detection identifies salient image features to guide autonomous robotic exploration.  There is little advance knowledge of the features in the scene or the proportion that should count as outliers.  A new algorithm addresses this ambiguity by modeling novel data in advance and characterizing regular data at run time.  Detection thresholds adapt dynamically to reduce misclassification risk while accommodating homogeneous and heterogeneous scenes.  Experiments demonstrate the technique on a representative set of navigation images from the Mars Exploration Rover "Opportunity."  An efficient image analysis procedure filters each image using the integral transform.  Pixel-level features are aggregated into covariance descriptors that represent larger regions.  Finally, a distance metric derived from generalized eigenvalues permits novelty detection with kernel density estimation.  Results suggest that exploiting training examples of novel data can improve performance in this domain.

Road network information simplifies autonomous driving by providing strong priors about environments. It informs a robotic vehicle with where it can drive, models of what can be expected, and contextual cues that influence driving behaviors. Currently, however, road network information is manually generated using a combination of GPS survey and aerial imagery. These manual techniques are labor intensive and error prone. To full exploit the benefits of digital imagery, these processes should be automated. As a step toward this goal, we present an algorithm that extracts the structure of parking lot visible from a given aerial image. To minimize human intervention in the use of aerial imagery, we devise a self-supervised learning algorithm that automatically generates a set of parking spot templates to learn the appearance of a parking lot and estimates the structure of the parking lot from the learned model. The data set extracted from a single image alone is too small to sufficiently learn an accurate parking spot model. However, strong priors trained using large data sets collected across multiple images dramatically improvce performance. Our self-supervised approach outperforms the prior alone by adapting the distribution of examples toward that found in the current image. A thorough empirical analysis compares leading state-of-the-art learning techniques on this problem.

Question-answer pairs extracted from email threads can help construct summaries of the thread, as well as inform semantic-based assistance with email. Previous work dedicated to email threads extracts only questions in interrogative form. We extend the scope of question and answer detection and pairing to encompass also questions in imperative and declarative forms, and to operate at sentence-level fidelity. Building on prior work, our methods are based on learned models over a set of features that include the content, context, and structure of email threads. For two large email corpora, we show that our methods balance precision and recall in extracting question-answer pairs, while maintaining a modest computation time.

Dialogue-based Question Answering (QA) is a highly complex task that brings together a QA system including various natural language processing components (i.e., components for question classification, information extraction, and retrieval) with dialogue systems for effective and natural communication. The dialogue-based access is difficult to establish when the QA system in use is complex and combines many different answer services with different quality and access characteristics. For example, some questions are processed by opendomain QA services with a broad coverage. Others should be processed by using a domain-specific instance ontology for more reliable answers. Different answer services may change their characteristics over time and the dialogue reaction models have to be updated according to that. To solve this problem, we developed introspective methods to integrate adaptable models of the answer services. We evaluated the impact of the learned models on the dialogue performance, i.e., whether the adaptable models can be used for a more convenient dialogue formulation process. We show significant effectiveness improvements in the resulting dialogues when using the machine learning (ML) models. Examples are provided in the context of the generation of system-initiative feedback to user questions and answers, as provided by heterogeneous information services.

Assigning significance in high-dimensional regression is challenging. Most computationally efficient selection algorithms cannot guard against inclusion of noise variables. Asymptotically valid p-values are not available. An exception is a recent proposal by Wasserman and Roeder (2008) which splits the data into two parts. The number of variables is then reduced to a manageable size using the first split, while classical variable selection techniques can be applied to the remaining variables, using the data from the second split. This yields asymptotic error control under minimal conditions. It involves, however, a one-time random split of the data. Results are sensitive to this arbitrary choice: it amounts to a `p-value lottery' and makes it difficult to reproduce results. Here, we show that inference across multiple random splits can be aggregated, while keeping asymptotic control over the inclusion of noise variables. We show that the resulting p-values can be used for control of both family-wise error (FWER) and false discovery rate (FDR). In addition, the proposed aggregation is shown to improve power while reducing the number of falsely selected variables substantially.

Many regression problems involve not one but several response variables (y's). Often the responses are suspected to share a common underlying structure, in which case it may be advantageous to share information across them; this is known as multitask learning. As a special case, we can use multiple responses to better identify shared predictive features -- a project we might call multitask feature selection. This thesis is organized as follows. Section 1 introduces feature selection for regression, focusing on ell_0 regularization methods and their interpretation within a Minimum Description Length (MDL) framework. Section 2 proposes a novel extension of MDL feature selection to the multitask setting. The approach, called the "Multiple Inclusion Criterion" (MIC), is designed to borrow information across regression tasks by more easily selecting features that are associated with multiple responses. We show in experiments on synthetic and real biological data sets that MIC can reduce prediction error in settings where features are at least partially shared across responses. Section 3 surveys hypothesis testing by regression with a single response, focusing on the parallel between the standard Bonferroni correction and an MDL approach. Mirroring the ideas in Section 2, Section 4 proposes a novel MIC approach to hypothesis testing with multiple responses and shows that on synthetic data with significant sharing of features across responses, MIC sometimes outperforms standard FDR-controlling methods in terms of finding true positives for a given level of false positives. Section 5 concludes.

Empirical testing is a very popular evaluation method for the development of intelligent systems. Here, previously solved problems with correct solutions are given as cases to the system. Validity is tested by comparing the expected results with the derived solutions. Besides classic forms of boolean testing of occurring solutions more refined methods are required for a thorough evaluation of real world knowledge systems. We present extended precision and recall functions for interactive knowledge systems that are generalizations of the existing measures. Additionally, we propose a visualization method for inspecting the validation result for interactive systems. A case study with a second-opinion system from the medical domain demonstrates the usefulness of the approach.